Norsk English

Green AI Deployment on Resource-Constrained Devices: Software Engineering Approaches for Edge Inference and Intermittent Computing

This thesis endeavors to explore software engineering techniques tailored for deploying AI models on devices like the Raspberry PI 5 and SEEED studio from Nordic semiconductor. The research will integrate concepts from intermittent computing and leverage for efficient edge inference, aiming to realize Green AI on resource-constrained devices.

Contact persons

AI-generated image

Master Project Description

With the proliferation of AI-driven applications, there is a growing demand to move AI to the edge, deploying models on resource-constrained devices. However, deploying energy-efficient AI on such devices remains a challenge. 

Research Topic Focus

  • Review the current challenges and strategies in deploying AI models on resource-constrained devices.
  • Investigate intermittent computing methodologies to optimize power and computational efficiency during AI model deployment.
  • Explore and implement software engineering practices tailored for Raspberry PI 5 and SEEED studio devices.
  • Optimize edge inference, assessing its compatibility and efficiency on the chosen platforms.
  • Conduct a series of experiments to evaluate the performance, power consumption, and efficiency of deployed AI models.

Expected Results

  • A curated set of best practices for deploying AI on resource-constrained devices with optimized energy consumption.
  • Successful implementation and benchmarking of AI models on Raspberry PI 5 and SEEED studio using possibly intermittent computing.
  • Demonstrable improvements in terms of energy efficiency and computational cost compared to traditional deployment methodologies.

Learning Outcomes

  • Acquire in-depth knowledge about the intricacies of deploying AI on resource-constrained devices.
  • Gain hands-on experience with intermittent computing and edge inference techniques.
  • Develop expertise in optimizing software engineering practices for Green AI applications.
  • Understand the challenges and benefits of implementing Green AI on devices like Raspberry PI 5 and SEEED studio.

Qualifications

  • Strong foundation in AI, machine learning, and software engineering principles.
  • Proficiency in relevant programming languages and familiarity with edge devices like Raspberry PI 5 and SEEED studio.
  • An interest in sustainable and energy-efficient technologies.

References

  1. Schwartz, R., Dodge, J., Smith, N. A., & Etzioni, O. (2019). Green AI. arXiv preprint arXiv:1907.10597.
  2. Gu, L., Li, X., & Han, J. (2017). Sleepy pi: A ultra-low power energy harvesting platform for intermittent computing systems. In Proceedings of the 8th International Workshop on Real-World Embedded Wireless Systems and Networks (pp. 25-30).
  3. https://tinyml.mit.edu/
  4. https://wiki.seeedstudio.com/Seeeduino-XIAO-TinyML/
  5. https://www.modular.com/mojo

Contact persons/supervisors

Sagar Sen ( Arda Goknil (), Erik Johannes Husom ()